Compressing 16-bit TIFFs

[Thiol]

Hi all. I'm in the processing of archiving a large number of 16-bit TIFF files (microscopy images from a science lab). LZW compression helped reduce disk space usage with 8-bit TIFFs, but it's actually making the 16-bitt images larger. I realize this is common with noisy images. Is there a good way to compress 16-bit TIFF files in a lossless manner?

 

[Lovesong]

For science images (which is what I work for) there is no real use in capturing things higher than 8-bit. I'm assuming that you're looking at either confocal stacks or stills from a light/fluorescent scope. For most things the fluorescent dye you're using has a pretty limited dynamic range, which means that your files are essentially bigger than they have to be.

While there is no "lossless" way to convert 16bit files to 8-bit files, in your place I would just use PS or Image J and simply save the files as 8-bit. I promise you that you will not notice a difference.

 

[MisterMe]

Hi all. I'm in the processing of archiving a large number of 16-bit TIFF files (microscopy images from a science lab). LZW compression helped reduce disk space usage with 8-bit TIFFs, but it's actually making the 16-bitt images larger. I realize this is common with noisy images. Is there a good way to compress 16-bit TIFF files in a lossless manner?

Your images are scientific data. You cannot change the format of your images without losing information. You always want access to the original data. Throwing away information is a no-no in science. Run your processing filters on copies. My guess is that a 16 GB Flash drive will stores 1000's of your images. Buy a few Flash drives and archive your images in triplicate.

 

[Thiol]

For science images (which is what I work for) there is no real use in capturing things higher than 8-bit. I'm assuming that you're looking at either confocal stacks or stills from a light/fluorescent scope. For most things the fluorescent dye you're using has a pretty limited dynamic range, which means that your files are essentially bigger than they have to be.

While there is no "lossless" way to convert 16bit files to 8-bit files, in your place I would just use PS or Image J and simply save the files as 8-bit. I promise you that you will not notice a difference.

They are actually grayscale electron microscopy images. Although the CCD camera we use technically has the grayscale dynamic range to handle all 16-bit, I think you're right in saying that most of the information in our images is noise. I just always acquired more data than I needed. Now I'm kind of stuck with them for the reasons given by the next poster (the whole can't alter the original images principle). Oh decisions decisions.

 

[ChrisA]

They are actually grayscale electron microscopy images. Although the CCD camera we use technically has the grayscale dynamic range to handle all 16-bit, I think you're right in saying that most of the information in our images is noise. I just always acquired more data than I needed. Now I'm kind of stuck with them for the reasons given by the next poster (the whole can't alter the original images principle). Oh decisions decisions.

Had the same problem is astronomical CCD images. At first we had the problem of "What to do with terabytes of image data?" And then the price of disks went down to $100 per terabyte. Problem solved.

 

[Edge100]

For science images (which is what I work for) there is no real use in capturing things higher than 8-bit. I'm assuming that you're looking at either confocal stacks or stills from a light/fluorescent scope. For most things the fluorescent dye you're using has a pretty limited dynamic range, which means that your files are essentially bigger than they have to be.

While there is no "lossless" way to convert 16bit files to 8-bit files, in your place I would just use PS or Image J and simply save the files as 8-bit. I promise you that you will not notice a difference.

8-bit does not cut it for my scientific imaging. We do maximum intensity projections of 60-100 slice Z-stack confocal or widefiled images, followed by 3D blind deconvolution, followed by image analysis. We typically are looking for radiation or drug-induced nuclear foci, which we count as a measure of DNA repair capacity.

In this respect, 16-bit is absolutely necessary for working with the final images for quantification. First off, thresholding would be a nightmare with only 256 levels (vs. 65000+). Second, after deconvolving an 8-bit image, you can definitely see blocky pixels where the lack of gradiation between levels of grey is evident. You don't want to show this type of thing at a scientific meeting.

8-bit is fine for a lab meeting Powerpoint slide, but it doesn't cut it for quantitative analysis or for publication-quality figures. Remember, bit depth has nothing to do with dynamic range. You could have a 4-bit image with a 20 stop dynamic range (which would look blocky as hell), or a 16-bit image with a 2 stop dynamic range (which would look smooth). Bit depth is ALL about the gradiations between levels of grey.

 

[Edge100]

Your images are scientific data. You cannot change the format of your images without losing information. You always want access to the original data. Throwing away information is a no-no in science. Run your processing filters on copies. My guess is that a 16 GB Flash drive will stores 1000's of your images. Buy a few Flash drives and archive your images in triplicate.

+1 for this.

Don't edit the original data. You can mess around with the 2D projections (if you're doing 3D stacks), but never touch the original captures, which should be 16-bit (for most newer cameras; at least 12-bit, I imagine).

 

[Thiol]

+1 for this.

Don't edit the original data. You can mess around with the 2D projections (if you're doing 3D stacks), but never touch the original captures, which should be 16-bit (for most newer cameras; at least 12-bit, I imagine).

I think I agree.

Just to be sure, is LZW completely kosher? With the right software, is opening an LZW compressed file and an uncompressed file are the same thing?